Development of an Optimized SVM Framework using Frilled-Lizard Metaheuristic Algorithm for Enhanced Classification of Salmonella Infection and Healthy Poultry Conditions

Abstract

Salmonellosis is a significant zoonotic disease affecting poultry and poses serious threats to food security and public health, while accurate discrimination between healthy and diseased birds is foundational to effective poultry farm management. However, the performance of conventional Support Vector Machine (SVM) classifiers is highly dependent on appropriate hyperparameters selection, which often requires extensive manual tuning and may lead to suboptimal classification performance. To address this limitation, this study presents an optimized Support Vector Machine (SVM) framework in which the recently proposed Frilled-Lizard Optimization (FLO) algorithm is adopted to optimize the SVM hyperparameters for enhanced classification of Salmonella infection and healthy poultry conditions across variable decision thresholds. The Frilled-Lizard Optimization algorithm(FLO) is a bio-inspired metaheuristic modelled on the hunting and retreat habit of the frilled lizard with botanical name Chlamydosaurus kingii, was utilized to automatically fine-tune SVM hyperparameters including the kernel function type, penalty factor (C), and kernel coefficient (γ). The dataset comprised 1,400 poultry fecal images sourced from Kaggle, with Salmonella represented by 400 images and Healthy birds by 320 images. The Pre-processing stage include, normalization of image, Contrast Limited Adaptive Histogram Equalization (CLAHE) contrast enhancement, conversion of grayscale,and feature extraction of PCA. Performance of the model makes use of 10-fold cross-validation at four different decision threshold values: 0.3, 0.4, 0.5, and 0.51. The output of the FLO algorithm converged to an optimal RBF kernel configuration with C = 5.7242 and γ = 0.7922 after 30 iterations, resulting to fitness value of 0.9829. The FLO-SVM achieved accuracy of 97.42% for Salmonella and 97.36% for Healthy at threshold 0.51, with sensitivity values of 93.57% and 93.75% respectively. False positive rates were reduced to 1.50% and 1.57%. Execution times were in the range of 40.02 to 42.09 seconds, which is a reduction of approximately 45% compared to the traditional SVM. Sensitivity analysis of thresholds showed increased specificity and precision at higher thresholds with minor trade-offs in sensitivity. The results confirm FLO SVM as an efficient and computationally feasible framework for the detection of Salmonella and classification of healthy birds, offering practical potential for deployment in intelligent poultry health monitoring systems.